Claude in Microsoft Foundry Is Generally Available on Azure

Claude in Microsoft Foundry became generally available on June 29, 2026. Enterprises can now build with Anthropic’s Claude models through the Azure environment they already govern, using existing authentication, billing and compliance controls. For organizations that have standardized on Azure and Entra ID, it’s a meaningful expansion of model choice without a new vendor relationship.

What is Claude in Microsoft Foundry?

It gives enterprises access to Claude models directly through their Azure account, hosted on Azure infrastructure. Teams reuse the procurement, networking and governance they already have for other Foundry models instead of setting up new processes.

Microsoft’s argument is that most enterprise AI projects stall not on model quality but on everything around the model: procurement, governance, networking and data handling. This release is positioned as removing those barriers so pilots can become production systems.

How developers access Claude in Microsoft Foundry

Developers call Claude through the Messages API, with prompt caching, extended thinking and tool streaming. For autonomous agents, Foundry Agent Service uses Claude as the reasoning core for multi-step planning, tool use and task execution across enterprise systems.

• Authentication through Microsoft Entra ID and existing Azure RBAC.
• Inference runs inside Azure, with a choice of Global or US data zones.
• Anthropic operates the inference and remains the data processor and SLA provider.
• Billing appears as Claude Consumption Units (CCU) on the Azure bill, with MACC drawdown supported.

Built on the Microsoft, NVIDIA and Anthropic partnership

The release builds on the strategic partnership announced in November 2025. Anthropic committed to buy $30 billion of Azure compute, with the option of up to one gigawatt more on NVIDIA Grace Blackwell and Vera Rubin systems. NVIDIA and Microsoft committed to invest up to $10 billion and $5 billion in Anthropic, respectively. Claude in Foundry runs on NVIDIA Blackwell Ultra systems connected by InfiniBand networking.

Governance, security and cost controls

For sensitive workloads, zero data retention is available, so prompts and completions aren’t kept by Anthropic after the call completes. Foundry Control Plane runs continuous evaluations to check agent responses against expectations and can block responses that break rules before they reach users.

On cost, Microsoft’s model router can send each query to the most suitable Claude model; Microsoft says this can cut inference costs by up to 50%. Treat that as a vendor figure to validate against your own workloads.

Customers already in production

Microsoft named several customers. NVIDIA uses Claude-powered agents daily for internal technical work. Momentic uses Claude to verify software releases from plain-English test descriptions. Everstar, a nuclear technology company, said a safety analysis that would have taken 200 human days took one. Bolt, a fintech platform, cited sustained throughput and reliability.

Where it fits in Microsoft’s multi-model strategy

Microsoft has moved away from a single-model story. Azure OpenAI remains central, Microsoft ships its own MAI models, and Claude now sits alongside them. Foundry increasingly works as a marketplace where teams pick the right model for each job and manage them all under one governance layer. Microsoft has also committed to continuing Claude access across GitHub Copilot, Microsoft 365 Copilot and Copilot Studio; for example, Copilot Cowork runs on Anthropic models.

What it means for Microsoft 365 and Power Platform admins

Foundry Agent Service increasingly powers custom agents across Microsoft 365, so model choice in Foundry can eventually shape what’s available to line-of-business agent builders. Before piloting:

• Review Azure governance policies, since Entra ID and RBAC carry over directly.
• Choose the Global or US data zone that fits your compliance posture.
• Tell finance to expect Claude Consumption Units on the Azure bill.
• If you plan to use Claude-based agents, lock down networking first; see Azure AI Foundry private endpoints.

Getting started

1. Open the Anthropic model catalog in Foundry. Existing Azure subscribers don’t need a separate Anthropic account.
2. Pick a data zone before deploying anything to production.
3. Use Foundry Agent Service for agents rather than calling the Messages API directly, so planning and tool use are handled for you.
4. Set budget guardrails early so a pilot doesn’t scale into a department-wide bill unnoticed.

How it compares to other clouds

Claude was already available through Amazon Bedrock and Google Vertex AI, so Azure was the gap. With this release, Anthropic’s models are on all three major clouds. For Microsoft-centric organizations, the differentiator isn’t availability but integration: one place for billing, identity and governance across Claude, Azure OpenAI and Microsoft’s own models.

Bottom line

Claude in Microsoft Foundry removes most of the procurement and governance friction that kept Claude out of Azure-first organizations. Confirm your data zone, budget controls and network isolation, then pilot it like any other Foundry model.

Last reviewed: October 2026

Sources

• Claude in Microsoft Foundry is now generally available (Microsoft Azure Blog)
• Microsoft, NVIDIA and Anthropic announce strategic partnerships (Official Microsoft Blog)
• Claude and Microsoft Foundry (Anthropic)
• Anthropic models catalog in Microsoft Foundry
• What is Microsoft Foundry Agent Service? (Microsoft Learn)